llama: guard buffer iteration against released device buffers

release_device_weights()/release_device_buffers() leave a null buffer in
ctxs_bufs while the device memory is released for on-demand VRAM sharing.
The memory-breakdown / total_size / clear paths iterated these and called
ggml_backend_buffer_get_size()/get_type()/clear() on the null buffer, tripping
GGML_ASSERT(buffer) and aborting on shutdown of a cold model. Skip null buffers
in llama_model::memory_breakdown and llama_kv_cache::{memory_breakdown,
total_size,clear}.

Assisted-by: Claude

common/arg.cpp

@@ -2575,15 +2575,17 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_env("LLAMA_ARG_CPU_MOE"));
     add_opt(common_arg(
         {"-ncmoe", "--n-cpu-moe"}, "N",
-        "keep the Mixture of Experts (MoE) weights of the first N layers in the CPU",
-        [](common_params & params, int value) {
-            if (value < 0) {
+        "keep the Mixture of Experts (MoE) weights of the first N layers in the CPU; "
+        "fractional N offloads part of the boundary layer at tensor granularity",
+        [](common_params & params, const std::string & value) {
+            const double n = std::stod(value);
+            if (n < 0) {
                 throw std::invalid_argument("invalid value");
             }
-            for (int i = 0; i < value; ++i) {
+            for (const std::string & re : llm_ffn_exps_cpu_block_regexes(n)) {
                 // keep strings alive and avoid leaking memory by storing them in a static vector
                 static std::list<std::string> buft_overrides;
-                buft_overrides.push_back(llm_ffn_exps_block_regex(i));
+                buft_overrides.push_back(re);
                 params.tensor_buft_overrides.push_back({buft_overrides.back().c_str(), ggml_backend_cpu_buffer_type()});
             }
         }

common/common.h

@@ -1080,6 +1080,28 @@ inline llama_model_tensor_buft_override llm_ffn_exps_cpu_override() {
     return { LLM_FFN_EXPS_REGEX, ggml_backend_cpu_buffer_type() };
 }
 
+// ATSInfer-style tensor-granularity static placement of MoE expert weights.
+// Offloads the expert weights of the first floor(n) layers to the CPU, plus a
+// subset of the boundary layer's three expert tensors for the fractional part.
+// Expert tensors are dropped from the GPU in ascending performance-density
+// order (gate, then up), keeping the higher-value down_proj resident longest.
+inline std::vector<std::string> llm_ffn_exps_cpu_block_regexes(double n_cpu_moe) {
+    std::vector<std::string> regexes;
+    const int n_full = n_cpu_moe > 0 ? (int) n_cpu_moe : 0;
+    for (int i = 0; i < n_full; ++i) {
+        regexes.push_back(llm_ffn_exps_block_regex(i));
+    }
+    const int k = (int) ((n_cpu_moe - n_full) * 3.0 + 0.5);
+    if (k >= 3) {
+        regexes.push_back(llm_ffn_exps_block_regex(n_full));
+    } else if (k == 2) {
+        regexes.push_back(string_format("blk\\.%d\\.ffn_(gate|up)_(ch|)exps", n_full));
+    } else if (k == 1) {
+        regexes.push_back(string_format("blk\\.%d\\.ffn_gate_(ch|)exps", n_full));
+    }
+    return regexes;
+}
+
 //
 // training utils
 //

include/llama.h

@@ -557,6 +557,16 @@ extern "C" {
 
     LLAMA_API const struct llama_model * llama_get_model   (const struct llama_context * ctx);
     LLAMA_API           llama_memory_t   llama_get_memory  (const struct llama_context * ctx);
+
+    // On-demand device (VRAM) residency: free the model's GPU weight buffers (keeping a host
+    // shadow so no reload is needed) to hand VRAM to another model, and rebuild them on demand.
+    // Any subsequent llama_decode auto-restores. If evict_kv is true, the KV cache device buffers
+    // are also evicted to a host shadow (for when the KV would not fit alongside the other model),
+    // at the cost of a D2H/H2D copy of the live KV each cycle; otherwise the KV stays resident.
+    // Intended for time-sharing a single GPU between several always-loaded models.
+    LLAMA_API void llama_context_release_device(struct llama_context * ctx, bool evict_kv);
+    LLAMA_API void llama_context_restore_device(struct llama_context * ctx);
+    LLAMA_API bool llama_context_weights_resident(const struct llama_context * ctx);
     LLAMA_API  enum llama_pooling_type   llama_pooling_type(const struct llama_context * ctx); // TODO: rename to llama_get_pooling_type
 
     LLAMA_API const struct llama_vocab * llama_model_get_vocab(const struct llama_model * model);

src/llama-context.cpp

@@ -728,6 +728,34 @@ void llama_context::synchronize() {
     t_compute_start_us = 0;
 }
 
+void llama_context::release_device(bool evict_kv) {
+    if (!model.weights_resident() && !(evict_kv && !kv_device_evicted)) {
+        return;
+    }
+    // ensure no compute is in flight before freeing the device buffers
+    synchronize();
+    model.release_device_weights();
+    if (evict_kv && memory && !kv_device_evicted) {
+        memory->release_device_buffers();
+        kv_device_evicted = true;
+    }
+}
+
+void llama_context::restore_device() {
+    if (model.weights_resident() && !kv_device_evicted) {
+        return;
+    }
+    model.restore_device_weights();
+    if (kv_device_evicted && memory) {
+        memory->restore_device_buffers();
+        kv_device_evicted = false;
+    }
+    // make sure all weight/KV uploads have completed before any compute reads them
+    if (sched) {
+        ggml_backend_sched_synchronize(sched.get());
+    }
+}
+
 const llama_model & llama_context::get_model() const {
     return model;
 }
@@ -1705,6 +1733,12 @@ int llama_context::decode(const llama_batch & batch_inp) {
         return -1;
     }
 
+    // on-demand VRAM: if the weights were released while this model was idle, bring them back
+    // to the device before building the compute graph
+    if (!model.weights_resident()) {
+        restore_device();
+    }
+
     const auto & vocab   = model.vocab;
     const auto & hparams = model.hparams;
 
@@ -3674,6 +3708,18 @@ void llama_synchronize(llama_context * ctx) {
     ctx->synchronize();
 }
 
+void llama_context_release_device(llama_context * ctx, bool evict_kv) {
+    ctx->release_device(evict_kv);
+}
+
+void llama_context_restore_device(llama_context * ctx) {
+    ctx->restore_device();
+}
+
+bool llama_context_weights_resident(const llama_context * ctx) {
+    return ctx->get_model().weights_resident();
+}
+
 float * llama_get_logits(llama_context * ctx) {
     ctx->synchronize();
 

src/llama-context.h

@@ -57,6 +57,14 @@ struct llama_context {
 
     void synchronize();
 
+    // on-demand device (VRAM) residency: free / rebuild the model's GPU weight buffers so that
+    // VRAM can be time-shared with another model. release_device() synchronizes first; decode()
+    // auto-restores when it finds the weights have been released. If evict_kv is set, the KV cache
+    // device buffers are also evicted to a host shadow (for when the KV would not fit alongside the
+    // other model) - this adds a D2H/H2D copy of the live KV on each cycle.
+    void release_device(bool evict_kv = false);
+    void restore_device();
+
     const llama_model   & get_model()   const;
     const llama_cparams & get_cparams() const;
 
@@ -345,6 +353,9 @@ private:
 
     bool sched_need_reserve = true;
 
+    // on-demand device residency: true while the KV cache device buffers have been evicted to host
+    bool kv_device_evicted = false;
+
     ggml_backend_t backend_cpu = nullptr;
     std::vector<ggml_backend_ptr> backends;
 

src/llama-kv-cache-iswa.cpp

@@ -272,6 +272,17 @@ void llama_kv_cache_iswa::state_read(llama_io_read_i & io, llama_seq_id seq_id,
     kv_swa->state_read(io, seq_id, flags);
 }
 
+void llama_kv_cache_iswa::release_device_buffers() {
+    kv_base->release_device_buffers();
+    kv_swa->release_device_buffers();
+}
+
+bool llama_kv_cache_iswa::restore_device_buffers() {
+    bool ok = kv_base->restore_device_buffers();
+    ok = kv_swa->restore_device_buffers() && ok;
+    return ok;
+}
+
 llama_kv_cache * llama_kv_cache_iswa::get_base() const {
     return kv_base.get();
 }

src/llama-kv-cache-iswa.h

@@ -83,6 +83,9 @@ public:
     void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) const override;
     void state_read (llama_io_read_i  & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) override;
 
+    void release_device_buffers() override;
+    bool restore_device_buffers() override;
+
     //
     // llama_kv_cache_iswa specific API
     //

src/llama-kv-cache.cpp

@@ -371,7 +371,9 @@ void llama_kv_cache::clear(bool data) {
 
     if (data) {
         for (auto & [_, buf] : ctxs_bufs) {
-            ggml_backend_buffer_clear(buf.get(), 0);
+            if (buf) { // may be null if evicted for on-demand VRAM sharing
+                ggml_backend_buffer_clear(buf.get(), 0);
+            }
         }
     }
 }
@@ -681,6 +683,9 @@ llama_pos llama_kv_cache::seq_pos_max(llama_seq_id seq_id) const {
 std::map<ggml_backend_buffer_type_t, size_t> llama_kv_cache::memory_breakdown() const {
     std::map<ggml_backend_buffer_type_t, size_t> ret;
     for (const auto & [ctx, buf] : ctxs_bufs) {
+        if (!buf) { // may be null if evicted for on-demand VRAM sharing
+            continue;
+        }
         ggml_backend_buffer_type_t buft = ggml_backend_buffer_get_type(buf.get());
 
         if (hparams.no_alloc) {
@@ -1801,12 +1806,88 @@ size_t llama_kv_cache::total_size() const {
     size_t size = 0;
 
     for (const auto & [_, buf] : ctxs_bufs) {
-        size += ggml_backend_buffer_get_size(buf.get());
+        if (buf) { // may be null if evicted for on-demand VRAM sharing
+            size += ggml_backend_buffer_get_size(buf.get());
+        }
     }
 
     return size;
 }
 
+void llama_kv_cache::release_device_buffers() {
+    // NOTE: the caller must have synchronized the backend so nothing references these buffers.
+    // The KV cache is read-write, so its host shadow is (re)captured fresh on every release.
+    if (dev_released) {
+        return;
+    }
+    dev_shadows.assign(ctxs_bufs.size(), device_buffer_shadow{});
+    size_t freed = 0;
+    for (size_t i = 0; i < ctxs_bufs.size(); ++i) {
+        ggml_context * ctx = ctxs_bufs[i].first.get();
+        ggml_backend_buffer_t buf = ctxs_bufs[i].second.get();
+        if (buf == nullptr || ggml_backend_buffer_is_host(buf) || ggml_backend_buffer_get_size(buf) == 0) {
+            continue; // only real device (VRAM) buffers are evictable
+        }
+        auto & sh = dev_shadows[i];
+        sh.releasable = true;
+        sh.buft = ggml_backend_buffer_get_type(buf);
+
+        // capture live contents compactly in stable iteration order (skip views, which alias a base)
+        size_t total = 0;
+        for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
+            if (t->view_src == nullptr) { total += ggml_nbytes(t); }
+        }
+        sh.data.resize(total);
+        size_t off = 0;
+        for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
+            if (t->view_src != nullptr) { continue; }
+            const size_t n = ggml_nbytes(t);
+            ggml_backend_tensor_get(t, sh.data.data() + off, 0, n);
+            off += n;
+        }
+
+        freed += ggml_backend_buffer_get_size(buf);
+        ctxs_bufs[i].second.reset(); // free the device buffer
+        for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
+            t->buffer = nullptr;
+            t->data   = nullptr;
+        }
+    }
+    dev_released = true;
+    if (freed > 0) {
+        LLAMA_LOG_INFO("%s: released %.2f MiB of KV cache from device\n", __func__, freed / 1024.0 / 1024.0);
+    }
+}
+
+bool llama_kv_cache::restore_device_buffers() {
+    if (!dev_released) {
+        return true;
+    }
+    for (size_t i = 0; i < ctxs_bufs.size(); ++i) {
+        auto & sh = dev_shadows[i];
+        if (!sh.releasable) {
+            continue;
+        }
+        ggml_context * ctx = ctxs_bufs[i].first.get();
+        ggml_backend_buffer_t buf = ggml_backend_alloc_ctx_tensors_from_buft(ctx, sh.buft);
+        if (buf == nullptr) {
+            LLAMA_LOG_ERROR("%s: failed to reallocate KV cache device buffer (out of VRAM?)\n", __func__);
+            return false;
+        }
+        size_t off = 0;
+        for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
+            if (t->view_src != nullptr) { continue; }
+            const size_t n = ggml_nbytes(t);
+            ggml_backend_tensor_set(t, sh.data.data() + off, 0, n);
+            off += n;
+        }
+        ctxs_bufs[i].second.reset(buf);
+    }
+    dev_released = false;
+    dev_shadows.clear(); // recaptured on next release
+    return true;
+}
+
 size_t llama_kv_cache::size_k_bytes() const {
     size_t size_k_bytes = 0;
 

src/llama-kv-cache.h

@@ -149,6 +149,10 @@ public:
     void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) const override;
     void state_read (llama_io_read_i  & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) override;
 
+    // on-demand device (VRAM) residency (see llama_memory_i)
+    void release_device_buffers() override;
+    bool restore_device_buffers() override;
+
     //
     // llama_kv_cache specific API
     //
@@ -265,6 +269,16 @@ private:
     // ggml contexts for the KV cache along with the allocated backend buffers:
     std::vector<std::pair<ggml_context_ptr, ggml_backend_buffer_ptr>> ctxs_bufs;
 
+    // on-demand device eviction: host shadow of each device buffer's live contents (re-captured on
+    // every release since the KV is read-write), plus the buffer type needed to reallocate it.
+    struct device_buffer_shadow {
+        ggml_backend_buffer_type_t buft = nullptr;
+        bool releasable = false;      // device (VRAM) buffer that was freed
+        std::vector<uint8_t> data;    // captured tensor bytes (compact, iteration order)
+    };
+    std::vector<device_buffer_shadow> dev_shadows; // parallel to ctxs_bufs
+    bool dev_released = false;
+
     // the current index from where we start searching for a free slot in the ring buffer of KV cells (see find_slot())
     // note: this is not part of the KV state and it's only used to speed-up the find_slot() method
     std::vector<uint32_t> v_heads;

src/llama-memory-hybrid.cpp

@@ -201,6 +201,18 @@ void llama_memory_hybrid::state_read(llama_io_read_i & io, llama_seq_id seq_id,
     mem_recr->state_read(io, seq_id, flags);
 }
 
+void llama_memory_hybrid::release_device_buffers() {
+    // evict the attention KV (grows with context); the recurrent state uses the no-op default
+    mem_attn->release_device_buffers();
+    mem_recr->release_device_buffers();
+}
+
+bool llama_memory_hybrid::restore_device_buffers() {
+    bool ok = mem_attn->restore_device_buffers();
+    ok = mem_recr->restore_device_buffers() && ok;
+    return ok;
+}
+
 llama_kv_cache * llama_memory_hybrid::get_mem_attn() const {
     return mem_attn.get();
 }

src/llama-memory-hybrid.h

@@ -76,6 +76,9 @@ public:
     void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) const override;
     void state_read (llama_io_read_i  & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0)       override;
 
+    void release_device_buffers() override;
+    bool restore_device_buffers() override;
+
     //
     // llama_memory_hybrid specific API
     //

src/llama-memory.h

@@ -124,6 +124,16 @@ struct llama_memory_i {
 
     virtual void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) const = 0;
     virtual void state_read (llama_io_read_i  & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) = 0;
+
+    //
+    // on-demand device (VRAM) residency
+    //
+
+    // Free this memory's device (VRAM) buffers to a host shadow and rebuild them on demand, to
+    // hand VRAM to another model while keeping the cached data (no re-prefill). The caller must
+    // have synchronized the backend first. Default: no-op (the memory stays resident).
+    virtual void release_device_buffers() {}
+    virtual bool restore_device_buffers() { return true; }
 };
 
 using llama_memory_ptr = std::unique_ptr<llama_memory_i>;

src/llama-model.cpp

@@ -1015,6 +1015,16 @@ struct llama_model::impl {
     // contexts where the model tensors metadata is stored as well as the corresponding buffers:
     std::vector<std::pair<ggml_context_ptr, std::vector<ggml_backend_buffer_ptr>>> ctxs_bufs;
 
+    // on-demand device (VRAM) weight residency: per-ctxs_bufs slot describing whether the
+    // device weight buffer can be freed and rebuilt from a host shadow (see release/restore_device_weights)
+    struct weight_release_slot {
+        ggml_backend_buffer_type_t buft = nullptr; // buffer type to reallocate on restore
+        bool releasable = false;                   // single-buffer alloc-path device (VRAM) buffer
+        bool resident   = true;                    // currently allocated on device
+        std::vector<uint8_t> shadow;               // host copy, captured lazily on first release
+    };
+    std::vector<weight_release_slot> weight_release; // parallel to ctxs_bufs
+
     buft_list_t cpu_buft_list;
     std::map<ggml_backend_dev_t, buft_list_t> gpu_buft_list;
 
@@ -1608,6 +1618,23 @@ bool llama_model_base::load_tensors(llama_model_loader & ml) {
 
         pimpl->ctxs_bufs.emplace_back(std::move(ctx_ptr), std::move(bufs));
 
+        // record on-demand release metadata (parallel to ctxs_bufs). Only the single-buffer
+        // alloc path on a device (non-host) buffer can be freed and rebuilt from a host shadow;
+        // the mmap buffer_from_host_ptr path and host (CPU) buffers are left resident.
+        {
+            llama_model::impl::weight_release_slot slot;
+            const bool mmap_host_ptr_path = ml.use_mmap && use_mmap_buffer && buffer_from_host_ptr_supported && is_default_buft;
+            auto & last_bufs = pimpl->ctxs_bufs.back().second;
+            if (!mmap_host_ptr_path && last_bufs.size() == 1) {
+                ggml_backend_buffer_t b = last_bufs[0].get();
+                if (b != nullptr && !ggml_backend_buffer_is_host(b)) {
+                    slot.releasable = true;
+                    slot.buft = buft;
+                }
+            }
+            pimpl->weight_release.push_back(std::move(slot));
+        }
+
         ctx_buf_maps.emplace_back(ctx, buf_map);
     }
 
@@ -1662,6 +1689,97 @@ ggml_tensor * llama_model_base::create_tensor(llama_model_loader & ml, const LLM
         tn, ne, flags);
 }
 
+void llama_model::release_device_weights() const {
+    // NOTE: the caller is responsible for synchronizing the backend scheduler first, so that no
+    // compute is in flight referencing these buffers when they are freed.
+    size_t freed = 0;
+    for (size_t i = 0; i < pimpl->ctxs_bufs.size(); ++i) {
+        auto & slot = pimpl->weight_release[i];
+        if (!slot.releasable || !slot.resident) {
+            continue;
+        }
+        ggml_context * ctx = pimpl->ctxs_bufs[i].first.get();
+        ggml_backend_buffer_t buf = pimpl->ctxs_bufs[i].second[0].get();
+        const size_t sz = ggml_backend_buffer_get_size(buf);
+
+        // lazily capture a host shadow of the weights (read-only, so captured only once).
+        // Store tensor bytes compactly in stable iteration order (independent of buffer layout /
+        // alignment padding) so restore does not depend on the re-allocated offsets matching.
+        // View tensors alias their base, so they are skipped (restored implicitly via their base).
+        if (slot.shadow.empty()) {
+            size_t total = 0;
+            for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
+                if (t->view_src != nullptr) continue;
+                total += ggml_nbytes(t);
+            }
+            slot.shadow.resize(total);
+            size_t off = 0;
+            for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
+                if (t->view_src != nullptr) continue;
+                const size_t n = ggml_nbytes(t);
+                ggml_backend_tensor_get(t, slot.shadow.data() + off, 0, n);
+                off += n;
+            }
+        }
+
+        // free the device (VRAM) buffer and clear the now-dangling tensor pointers so that
+        // ggml_backend_alloc_ctx_tensors_from_buft reallocates them cleanly on restore
+        pimpl->ctxs_bufs[i].second[0].reset();
+        for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
+            t->buffer = nullptr;
+            t->data   = nullptr;
+        }
+        slot.resident = false;
+        freed += sz;
+    }
+    if (freed > 0) {
+        LLAMA_LOG_INFO("%s: released %.2f MiB of device weights\n", __func__, freed / 1024.0 / 1024.0);
+    }
+}
+
+bool llama_model::restore_device_weights() const {
+    size_t restored = 0;
+    for (size_t i = 0; i < pimpl->ctxs_bufs.size(); ++i) {
+        auto & slot = pimpl->weight_release[i];
+        if (!slot.releasable || slot.resident) {
+            continue;
+        }
+        ggml_context * ctx = pimpl->ctxs_bufs[i].first.get();
+        ggml_backend_buffer_t buf = ggml_backend_alloc_ctx_tensors_from_buft(ctx, slot.buft);
+        if (buf == nullptr) {
+            LLAMA_LOG_ERROR("%s: failed to reallocate device weight buffer (out of VRAM?)\n", __func__);
+            return false;
+        }
+        ggml_backend_buffer_set_usage(buf, GGML_BACKEND_BUFFER_USAGE_WEIGHTS);
+
+        // re-upload weights from the compact host shadow, using the same stable iteration order
+        // and view-skipping as the capture above
+        size_t off = 0;
+        for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
+            if (t->view_src != nullptr) continue;
+            const size_t n = ggml_nbytes(t);
+            ggml_backend_tensor_set(t, slot.shadow.data() + off, 0, n);
+            off += n;
+        }
+        pimpl->ctxs_bufs[i].second[0].reset(buf);
+        slot.resident = true;
+        restored += ggml_backend_buffer_get_size(buf);
+    }
+    if (restored > 0) {
+        LLAMA_LOG_INFO("%s: restored %.2f MiB of device weights\n", __func__, restored / 1024.0 / 1024.0);
+    }
+    return true;
+}
+
+bool llama_model::weights_resident() const {
+    for (const auto & slot : pimpl->weight_release) {
+        if (slot.releasable && !slot.resident) {
+            return false;
+        }
+    }
+    return true;
+}
+
 std::string llama_model::arch_name() const {
     return llm_arch_name(arch);
 }
@@ -1714,6 +1832,11 @@ std::map<ggml_backend_buffer_type_t, size_t> llama_model::memory_breakdown() con
             ret[buft] += ggml_backend_alloc_ctx_tensors_from_buft_size(ctx.get(), buft);
         } else {
             for (const auto & buf : bufs) {
+                // buf may be null if the device weights were released for on-demand VRAM sharing
+                // (see release_device_weights); skip it in the breakdown
+                if (!buf) {
+                    continue;
+                }
                 // GGML_ASSERT(ggml_backend_buffer_get_base(buf.get()) != nullptr); // multi_buffer does not have a defined base
                 ret[ggml_backend_buffer_get_type(buf.get())] += ggml_backend_buffer_get_size(buf.get());
             }

src/llama-model.h

@@ -663,6 +663,14 @@ struct llama_model {
 
     const struct ggml_tensor * get_tensor(const char * name) const;
 
+    // on-demand device (VRAM) weight residency: free the GPU weight buffers (keeping a host
+    // shadow) and rebuild them on demand. Used to time-share VRAM between models without
+    // reloading or losing the KV cache. release_device_weights() requires the caller to have
+    // synchronized the backend scheduler first.
+    void release_device_weights() const;
+    bool restore_device_weights() const;
+    bool weights_resident() const;
+
     float get_rope_freq_base (const llama_cparams & cparams, int il) const;
     float get_rope_freq_scale(const llama_cparams & cparams, int il) const;
 

tools/server/server-context.cpp

@@ -25,6 +25,8 @@
 #include <filesystem>
 #include <utility>
 #include <fstream>
+#include <thread>
+#include <atomic>
 
 // fix problem with std::min and std::max
 #if defined(_WIN32)
@@ -35,6 +37,16 @@
 #include <windows.h>
 #endif
 
+// POSIX file locking + inotify doorbell for the cross-process VRAM arbiter (see vram_share_* below)
+#if !defined(_WIN32)
+#include <sys/file.h>
+#include <sys/stat.h>
+#include <sys/inotify.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <poll.h>
+#endif
+
 using json = nlohmann::ordered_json;
 
 constexpr int HTTP_POLLING_SECONDS = 1;
@@ -871,6 +883,8 @@ public:
     }
 
     ~server_context_impl() {
+        // stop the VRAM warden thread (and release the token) before tearing anything down
+        vram_share_shutdown();
         if (!sleeping) {
             // destroy() is already called when entering sleeping state
             // we don't call it again here to avoid double free
@@ -894,6 +908,188 @@ private:
     llama_model   * model_dft = nullptr;
     llama_context * ctx_dft   = nullptr;
 
+    // Cross-process VRAM arbiter: when LLAMA_SLEEP_VRAM_ONLY is set, several always-loaded
+    // llama-server processes time-share one GPU. A single flock() on <arena>/token.lock is the
+    // baton: "resident (weights in VRAM) iff I hold the token". A model warms up (locks + restores
+    // weights) right before it decodes, and goes cold (releases weights + unlocks) when it idles,
+    // so only one model holds VRAM at a time and the KV cache is never evicted (no re-prefill).
+    bool vram_only    = false;   // release-mode sleep active (LLAMA_SLEEP_VRAM_ONLY)
+    bool vram_evict_kv = false;  // also evict the KV cache to host on release (LLAMA_SLEEP_EVICT_KV)
+    bool vram_flock   = false;   // cross-process flock coordination available
+    std::atomic<bool> vram_cold{true}; // weights currently released (read by warden thread)
+    int  vram_lock_fd = -1;      // fd for <arena>/token.lock
+
+    // inotify "doorbell": a waiter that wants the VRAM token touches <arena>/doorbell/<pid>, which
+    // wakes the current holder's warden thread so it releases immediately instead of only on idle.
+    std::string       vram_doorbell_dir;
+    std::string       vram_pid_str;
+    int               vram_inotify_fd = -1;
+    std::thread       vram_warden;
+    std::atomic<bool> vram_warden_run{false};
+
+    // open the shared arena (flock token + doorbell dir). Idempotent; sets vram_only/vram_flock.
+    void vram_arena_open() {
+        if (getenv("LLAMA_SLEEP_VRAM_ONLY") == nullptr) {
+            return;
+        }
+        vram_only = true;
+        vram_evict_kv = getenv("LLAMA_SLEEP_EVICT_KV") != nullptr;
+#if !defined(_WIN32)
+        if (vram_lock_fd >= 0) {
+            return; // already open
+        }
+        const char * arena_env = getenv("LLAMA_VRAM_ARENA");
+        const std::string arena = arena_env ? arena_env : "/dev/shm/llama-vram";
+        mkdir(arena.c_str(), 0777);
+        const std::string lock_path = arena + "/token.lock";
+        vram_lock_fd = open(lock_path.c_str(), O_RDWR | O_CREAT, 0666);
+        if (vram_lock_fd >= 0) {
+            vram_flock = true;
+            vram_pid_str = std::to_string(getpid());
+            vram_doorbell_dir = arena + "/doorbell";
+            mkdir(vram_doorbell_dir.c_str(), 0777);
+            SRV_INF("VRAM arbiter: cross-process GPU sharing via %s (doorbell %s)\n",
+                    lock_path.c_str(), vram_doorbell_dir.c_str());
+        } else {
+            SRV_WRN("VRAM arbiter: cannot open %s, running VRAM-only sleep without cross-process lock\n", lock_path.c_str());
+        }
+#endif
+    }
+
+    // Acquire the VRAM token BEFORE uploading this model's weights, so any model currently resident
+    // on the GPU releases first and the load uploads into free VRAM instead of racing it (which
+    // could OOM, e.g. the 4B task model holding VRAM while a large model loads). Blocks until free.
+    // Called from load_model() right before common_init_from_params().
+    void vram_acquire_for_load() {
+        vram_arena_open();
+        if (!vram_only) {
+            return;
+        }
+#if !defined(_WIN32)
+        if (vram_flock) {
+            vram_ring_doorbell();         // nudge whoever is resident to release
+            flock(vram_lock_fd, LOCK_EX); // block until the GPU is free
+        }
+#endif
+        vram_cold = false; // we hold the token; weights will be resident after the upload
+    }
+
+    // Start the doorbell warden. Called from init() after the (coordinated) load. The model stays
+    // warm (holding the token acquired in vram_acquire_for_load) and serves its first request
+    // without a re-warm; the warden releases it when another model rings the doorbell.
+    void vram_share_init() {
+        vram_arena_open();
+        if (!vram_only) {
+            return;
+        }
+        vram_cold = false; // resident and holding the token after the coordinated load
+#if !defined(_WIN32)
+        if (vram_flock && vram_inotify_fd < 0) {
+            vram_inotify_fd = inotify_init1(IN_NONBLOCK);
+            if (vram_inotify_fd >= 0) {
+                inotify_add_watch(vram_inotify_fd, vram_doorbell_dir.c_str(), IN_CLOSE_WRITE);
+                vram_warden_run = true;
+                vram_warden = std::thread([this]{ vram_warden_loop(); });
+            }
+        }
+#endif
+    }
+
+    // warden thread: block on the doorbell; when another process rings (wants the token) and we
+    // currently hold it, ask the loop to yield (release) at its next idle point. Only touches the
+    // thread-safe queue - never the GPU - so it cannot race a decode.
+    void vram_warden_loop() {
+#if !defined(_WIN32)
+        char buf[4096];
+        while (vram_warden_run.load()) {
+            struct pollfd pfd { vram_inotify_fd, POLLIN, 0 };
+            int pr = poll(&pfd, 1, 500); // 500ms so we periodically re-check the run flag
+            if (pr <= 0) {
+                continue;
+            }
+            ssize_t n = read(vram_inotify_fd, buf, sizeof(buf));
+            if (n <= 0) {
+                continue;
+            }
+            bool foreign_ring = false;
+            for (char * p = buf; p < buf + n; ) {
+                struct inotify_event * ev = (struct inotify_event *) p;
+                if (ev->len > 0 && vram_pid_str != ev->name) {
+                    foreign_ring = true; // someone else wants the GPU
+                }
+                p += sizeof(struct inotify_event) + ev->len;
+            }
+            if (foreign_ring && !vram_cold) {
+                queue_tasks.request_yield();
+            }
+        }
+#endif
+    }
+
+    // ring the doorbell so the current token holder releases promptly
+    void vram_ring_doorbell() {
+#if !defined(_WIN32)
+        if (!vram_flock || vram_doorbell_dir.empty()) {
+            return;
+        }
+        const std::string f = vram_doorbell_dir + "/" + vram_pid_str;
+        int fd = open(f.c_str(), O_WRONLY | O_CREAT | O_TRUNC, 0666);
+        if (fd >= 0) {
+            ssize_t w = write(fd, "1", 1);
+            (void) w;
+            close(fd);
+        }
+#endif
+    }
+
+    // acquire the VRAM token (blocking) and bring weights back to the device. Called on the loop
+    // thread right before a decode, so it never races compute.
+    void vram_ensure_warm() {
+        if (!vram_only || !vram_cold) {
+            return;
+        }
+#if !defined(_WIN32)
+        if (vram_flock) {
+            vram_ring_doorbell();         // nudge the current holder to release
+            flock(vram_lock_fd, LOCK_EX); // blocks until the current holder goes cold
+        }
+#endif
+        llama_context_restore_device(ctx_tgt);
+        if (ctx_dft != nullptr) {
+            llama_context_restore_device(ctx_dft);
+        }
+        vram_cold = false;
+    }
+
+    void vram_share_shutdown() {
+#if !defined(_WIN32)
+        vram_warden_run = false;
+        if (vram_warden.joinable()) {
+            vram_warden.join();
+        }
+        if (vram_inotify_fd >= 0) { close(vram_inotify_fd); vram_inotify_fd = -1; }
+        if (vram_lock_fd    >= 0) { flock(vram_lock_fd, LOCK_UN); close(vram_lock_fd); vram_lock_fd = -1; }
+#endif
+    }
+
+    // release the device weights (keeping KV + host shadow) and drop the VRAM token so another
+    // model can warm up. Called on the loop thread when the server goes idle.
+    void vram_go_cold() {
+        if (!vram_only || vram_cold) {
+            return;
+        }
+        llama_context_release_device(ctx_tgt, vram_evict_kv);
+        if (ctx_dft != nullptr) {
+            llama_context_release_device(ctx_dft, vram_evict_kv);
+        }
+#if !defined(_WIN32)
+        if (vram_flock) {
+            flock(vram_lock_fd, LOCK_UN);
+        }
+#endif
+        vram_cold = true;
+    }
+
     common_speculative_init_result_ptr spec_init;
 
     common_context_seq_rm_type ctx_tgt_seq_rm_type = COMMON_CONTEXT_SEQ_RM_TYPE_NO;
@@ -951,6 +1147,25 @@ private:
 
     void handle_sleeping_state(bool new_state) {
         GGML_ASSERT(sleeping != new_state);
+
+        // Lightweight VRAM-only sleep: instead of a full unload/reload (which frees RAM and the
+        // KV cache and pays a full reload on wake), just release the model's device (VRAM) weight
+        // buffers to a host shadow (keeping the context, KV cache and host weights) and drop the
+        // shared VRAM token. Waking is handled lazily by vram_ensure_warm() right before the next
+        // decode (which re-acquires the token first), so a single GPU is time-shared between models
+        // with no re-prefill.
+        if (vram_only && ctx_tgt != nullptr) {
+            if (new_state) {
+                SRV_INF("%s", "server entering sleeping state (VRAM-only: releasing device weights, keeping KV cache)\n");
+                vram_go_cold();
+            } else {
+                SRV_INF("%s", "server exiting sleeping state (VRAM-only: weights restored on next decode)\n");
+                // token is acquired and weights restored by vram_ensure_warm() before decode
+            }
+            sleeping = new_state;
+            return;
+        }
+
         if (new_state) {
             SRV_INF("%s", "server is entering sleeping state\n");
             destroy();
@@ -1142,6 +1357,11 @@ private:
             params_base.load_progress_callback_user_data = &load_progress_text;
         }
 
+        // VRAM arbiter: acquire the GPU token before uploading weights, so any resident model (e.g.
+        // the warm 4B task model) releases first and this load uploads into free VRAM instead of
+        // racing it and OOM-ing. No-op unless LLAMA_SLEEP_VRAM_ONLY is set.
+        vram_acquire_for_load();
+
         llama_init = common_init_from_params(params_base);
 
         model_tgt = llama_init->model();
@@ -1409,6 +1629,11 @@ private:
             handle_sleeping_state(sleeping);
         });
 
+        // VRAM arbiter: start the doorbell warden. The load was coordinated (vram_acquire_for_load
+        // grabbed the token before uploading), so we stay warm holding the token and serve the first
+        // request without a re-warm; the warden releases us when another model rings the doorbell.
+        vram_share_init();
+
         metrics.init();
 
         if (params_base.cache_idle_slots) {
@@ -3589,6 +3814,10 @@ private:
             n_empty_consecutive = 0;
         }
 
+        // VRAM arbiter: make sure our weights are resident (acquiring the shared GPU token first)
+        // before we decode. Runs on the loop thread, so it never races an in-flight decode.
+        vram_ensure_warm();
+
         const int ret = llama_decode(ctx_tgt, batch_view);
 
         metrics.on_decoded(slots);
@@ -4010,8 +4239,12 @@ struct server_res_generator : server_res_spipe {
     server_response_reader rd;
     server_res_generator(server_queue & queue_tasks, server_response & queue_results, int sleep_idle_seconds, bool bypass_sleep = false)
             : rd(queue_tasks, queue_results, HTTP_POLLING_SECONDS) {
-        // fast path in case sleeping is disabled
-        bypass_sleep |= sleep_idle_seconds < 0;
+        // fast path in case sleeping is disabled. Note: the VRAM arbiter (LLAMA_SLEEP_VRAM_ONLY)
+        // can put the server to sleep via the cross-process doorbell even when idle-sleep is
+        // disabled (sleep_idle_seconds < 0), so in that case requests must still wake it - otherwise
+        // a doorbell-slept server would hang, never returning from a request.
+        static const bool vram_arbiter = getenv("LLAMA_SLEEP_VRAM_ONLY") != nullptr;
+        bypass_sleep |= (sleep_idle_seconds < 0 && !vram_arbiter);
         if (!bypass_sleep) {
             queue_tasks.wait_until_no_sleep();
         }

tools/server/server-queue.cpp

@@ -116,6 +116,12 @@ void server_queue::wait_until_no_sleep() {
     }
 }
 
+void server_queue::request_yield() {
+    std::unique_lock<std::mutex> lock(mutex_tasks);
+    yield_requested = true;
+    condition_tasks.notify_all();
+}
+
 void server_queue::terminate() {
     std::unique_lock<std::mutex> lock(mutex_tasks);
     running = false;
@@ -129,6 +135,9 @@ void server_queue::start_loop(int64_t idle_sleep_ms) {
     constexpr auto max_wait_time = std::chrono::seconds(1);
     auto should_sleep = [&]() -> bool {
         // caller must hold mutex_tasks
+        if (yield_requested) {
+            return true; // another process rang the VRAM doorbell - release now
+        }
         if (idle_sleep_ms < 0) {
             return false;
         }
@@ -178,6 +187,7 @@ void server_queue::start_loop(int64_t idle_sleep_ms) {
             if (should_sleep()) {
                 QUE_INF("%s", "entering sleeping state\n");
                 sleeping = true;
+                yield_requested = false; // consumed
                 callback_sleeping_state(true);
                 req_stop_sleeping = false;
                 // wait until we are requested to exit sleeping state
@@ -195,13 +205,17 @@ void server_queue::start_loop(int64_t idle_sleep_ms) {
                 condition_tasks.notify_all(); // notify wait_until_no_sleep()
                 break; // process new tasks
             } else {
-                // wait for new tasks or timeout for checking sleeping condition
+                // wait for new tasks, a VRAM yield request, or timeout for checking sleeping condition
                 bool res = condition_tasks.wait_for(lock, max_wait_time, [&]{
-                    return (!queue_tasks.empty() || !running);
+                    return (!queue_tasks.empty() || !running || yield_requested);
                 });
-                if (res) {
+                if (res && !queue_tasks.empty()) {
                     break; // new task arrived or terminate
                 }
+                if (!running) {
+                    break;
+                }
+                // otherwise (timeout or yield request), loop again to re-check should_sleep
                 // otherwise, loop again to check sleeping condition
             }
         }

tools/server/server-queue.h

@@ -16,6 +16,7 @@ private:
     bool running  = false;
     bool sleeping = false;
     bool req_stop_sleeping = false;
+    bool yield_requested = false; // set by request_yield() when another process wants the VRAM token
     int64_t time_last_task = 0;
 
     // queues
@@ -51,6 +52,11 @@ public:
     // returns immediately if not sleeping
     void wait_until_no_sleep();
 
+    // request that the loop go to sleep (release VRAM) as soon as it is idle - called from the
+    // VRAM-arbiter warden thread when another process rings the doorbell for the GPU token.
+    // Thread-safe; wakes the loop so it releases promptly instead of at the next idle poll.
+    void request_yield();
+
     bool is_sleeping() {
         std::unique_lock<std::mutex> lock(mutex_tasks);
         return sleeping;